San Francisco, September 20, 2026: Many experts are calling AI (Artificial Intelligence) a threat to human culture. Meanwhile, incidents are emerging that indicate that AI is becoming uncontrolled. During a test in May, Google's AI 'Gemini' exceeded its limits and infiltrated the networks of three companies.
According to Google, these incidents occurred during tests conducted to evaluate the cyber security capabilities of Gemini. After entering the networks of the three companies, the AI model stopped the attacks on its own.
Even before these hacking incidents came to light, similar incidents had already happened at other leading AI labs like Anthropic, OpenAI and Meta, raising fears that this technology could go beyond human control.
In response, some industry leaders, such as Dario Amodei, CEO of Anthropic, have advocated slowing down the pace of AI development. On the other hand, others, such as Jensen Huang, CEO of chip maker Nvidia, argue that AI development should continue rapidly.
Gemini attacked real company instead of dummy company
Google said that in each case, its AI model was instructed to attack a dummy company. However, the name of the dummy company used in the test resembled the name of a real company. When Gemini got access to the Internet, it started trying to penetrate the security system of the real company.
AI model logged in using passwords and guesses available on the internet
The Gemini model used passwords that were either obtained online or learned through guesswork to log into the online infrastructure of the target company and two other companies.
Google said that after logging in, the Gemini model realized that it was accessing real companies' infrastructure rather than the simulated environment that was part of the testing, and it stopped the attack. Google also said that the companies that were hacked due to its technology did not suffer any loss.
Read more: Massive fire in Volvo car crashing into divider in Telangana, tragic death of renowned neurosurgeon